2014

Lab Manual on Soft Computing

[IT-802]

Ms. Neha Sexana

INSTITUTE OF INFORMATION
TECHNOLOGY &
MANAGEMENT , GWALIOR
01-Feb-2014
SOFTWARE REQUIREMENT :
1. Turbo C++ IDE (TurboC3)
2. Borland Turbo C++ (Version 4.5)
3. JAVA

REFERENCE BOOKS :
1. S.N. Shivnandam, “Principle of soft computing”, Wiley.

2. S. Rajshekaran and G.A.V. Pai, “Neural Network , Fuzzy logic

And Genetic Algorithm”, PHI.

3. Jack M. Zurada, “Introduction to Artificial Neural Network

System” JAico Publication.

4. Simon Haykins, “Neural Network- A Comprehensive

Foudation”
 LIST OF PROGRAMS

1. Write A Program For Implementing Linear Saturating

Function.

2. Study And Analysis Of Art Model.

3. Write A Program For Error Back Propagation Algorithm

(Ebpa) Learning.

4. Study And Analysis Of CPN

5. Study And Analysis Of Genetic Algorithm Life Cycle.

6. Study And Analysis Of Fuzzy Vs Crisp Logic.

7. Write A Program Of Perceptron Training Algorithm.

8. Write A Program To Implement Hebb’s Rule

9. Write A Program To Implement Of Delta Rule

10. Write A Program For Back Propagation Algorithm

11. Write A Program To Implement Logic Gates

 EXPERIMENT: 1

DESCRIPTION: WAP FOR IMPLEMENTING LINEAR SATURATING

FUNCTION.

#include<iostream.h>
#include<conio.h>
void main()
{
clrscr();
float x,b,w,net;
float out;
cout<<"enter the input X=";
cin>>x;
cout<<"enter the bias b=";
cin>>b;
cout<<"enter the weight W=";
cin>>w;
net=(w*x+b);
cout<<"net="<<net<<endl;
if(net<0)
{
out=0;
}
else
if((net>=0)&&(net<=1))
{
out=net;
}
else
out=1;
cout<<"output="<<out<<endl;
getch();
}
 OUTPUT

When, net<0 :-

When, ((net>=0)&&( net<=1)) :-

When, net>1 :-
 EXPERIMENT: 2

DESCRIPTION: STUDY AND ANALYSIS OF ART MODEL.

Adaptive Resonance Theory (ART) :-

ART (Adaptive Resonance Theory) models are neural networks that

performs clustering and can allow the no. of cluster to vary with problem size.
The difference between ART and other clustering model is that ART
allows the user to control the degree of similarity between no. of same cluster by
means of a user define constants called vigilance parameter.
The 1st version of ART was ART1, developed by Carpenter and
Grossberg in 1988.

Block Diagram of ART ->

It consist of two layers of neurons labeled “Comparision” and

“Recognition”. Gain1, Gain2 and reset provide control functions needed for training
and classification.
In a real world, the network will be exposed to constantly changing
environment i.e. it may never see the same training vector twice, in such cases any of
the conventional ANN seems till now will learn nothing.
ART network is a vector classifier. It accept the inputs and classifies it
on the basis of already stored patterns to which it most resembles. The classification is
done by the recognition layer.
If the input vector does not matches any already stored pattern, a new
pattern or category is created which matches the input vector.
No stored pattern is ever modified, if it does not match the current
input within the vigilance.

1. Comparision Layer -> It works on 2/3 rule. This layer recieves the binary layer
input vector represented by X and initially passes it trough unchanged to become
the vector C.
For one iteration C=X, in later case binary vector R is
produce from Recognition layer. Each neuron in the comparision layer recieves
three binary inputs.
a. The component Xi from the input vector X.
b. The feedback signal Pj i.e. the weighted sum of recognition layer output.
c. The input from the gain signal i.e. Gain1 (G1).
 OR of X OR of R G2
Component Component
0 0 0
1 0 1
1 1 0
0 1 0

2. Recognition Layer -> The recognition layer serves to classify the input vector.
Each recognition layer neuron has an associated weight vector bj only the neuron
with the weight vector best matching the input vectors, fire all other neurons are
inhibited.
The weights in the recognition layer make a stored pattern
for a category of input output vector, these weights are real number.
The binary version of same pattern is stored in the
corresponding set of weights in the comparision layer.

Operation perform in Recognition Layer -> Each recognition layer neuron

computes a dot(.) product between its weights and incoming vector C. The
neuron which has the highest dot product will win the competition and will fire,
the response of recognition layer is in the fashion “winner takes all”.

a. GAIN2 (G2) -> It is set to 1 if one of the component of input vector X is 1. G2

is logically “OR” of the component of X.

b. GAIN1 (G1) -> Like GAIN2, the output of GAIN1 is 1. If any component of
binary input vector is 1.
But if any component of R is 1 then, G1 is force to become 0.
c. Reset Signal -> The Reset modules measures the similarity between the vector
X and C. Generally, this similarity is the ratio of 1’s in the vector X.
If this ratio is below the vigilance parameter, a reset signal is
issue and inhibits the neuron of recognition layer.
The vigilance parameter p is closer to 1 for accuracy & its
value is 0.9
 EXPERIMENT: 3

DESCRIPTION: WAP FOR ERROR BACK PROPAGATION ALGORITHM

(EBPA) LEARNING.

#include<conio.h>
#include<iostream.h>
#include<math.h>
void main()
{
clrscr();
float l,c,s1,n1,n2,w10,b10,w20,b20,w11,b11,w21,b21,p,t,a0=-1,a1,a2,e,s2;
cout<<"enter the input weights/base of second n/w= ";
cin>>w10>>b10;
cout<<"enter the input weights/base of second n/w= ";
cin>>w20>>b20;
cout<<"enter the learning coefficient of n/w c= ";
cin>>c;
/* Step1:Propagation of signal through n/w */
n1=w10*p+b10;
a1=tanh(n1);
n2=w20*a1+b20;
a2=tanh(n2);
e=(t-a2);
/* Back Propagation of Sensitivities */
s2=-2*(1-a2*a2)*e;
s1=(1-a1*a1)*w20*s2;
/* Updation of weights and bases */
w21=w20-(c*s2*a1);
w11=w10-(c*s1*a0);
b21=b20-(c*s2);
b11=b10-(c*s1);
cout<<"The uploaded weight of first n/w w11= "<<w11;
cout<<"\n"<<"The uploaded weight of second n/w w21= "<<w21;
cout<<"\n"<<"The uploaded base of second n/w b11= "<<b11;
cout<<"\n"<<"The uploaded base of second n/w b21= "<<b21;
getch();
}
OUTPUT
 EXPERIMENT: 4

DESCRIPTION: STUDY AND ANALYSIS OF CPN.

Counter Propagation Network (CPN) :-

Counter Propagation Network (CPN) is not as general as Back

Propagation. But it provides a solution for those application that cannot tolerate long
training time.
It is a combination of two well known algorithm i.e. Self Organizing
Map of Kohonen and Grossberg outstar. The training process associates the input
vector with the corresponding output vector.
These vectors may be binary or continuous. It has the generalization
capability. The generalization capability of network allows it to produce a correct
output even when it is given an input vector i.e. partially incomplete or partially
incorrect.
CPN is used to compress the data before sending and decompress the
data before receiving. With the help of these network we can compress audio data
upto 100:1 and video data upto 10:1.

Architecture Of CPN ->

It consist of one input layer, one Kohonen layer and one Grossberg
layer.
All the units of input layer are fully interconnected by weights to the
units of Kohonen layer.
Similarly, all the units of Kohonen layer are fully interconnected by
weights to the Grossberg layer.
It works in two mode.

1. Normal Mode ->

When we apply the input vector and get the desired output.

2. Training Mode ->

Apply input vector and modify the network weights so that, we
get the desired output.
Advantages of CPN ->

1. It is simple and it forms a good statistical model of its input vector

environment.

2. CPN trains rapidly. If appropriately applied, it can safe large amount of

computer time.

3. It is also useful for rapid prototyping of system, where the greater accuracy of
Back Propagation makes it the method of choice in the final version. But a
quick approximation is important so, CPN is more useful then Back
Propagation.
 EXPERIMENT: 5

DESCRIPTION: STUDY AND ANALYSIS OF GENETIC ALGORITHM LIFE

CYCLE.

Genetic Algorithm (GA) :-

Genetic Algorithm is a search heuristic (experience) that follows the

process of natural evolution.
This heuristic is used to generate useful solutions to optimization and
search problems.
Genetic Algorithm belong to the larger class of evolutionary
algorithm (EA) which generate solutions to optimization problems and using
techniques inspired by natural evolution like – inheritance, mutation, selection,
crossover.
Genetic Algorithm need design space to be converted into genetic
space. Genetic Algorithm works with coding variables.
Genetic Algorithm uses population of point at one time in contrast to
the single point approach. It means that genetic algorithm processes a number of
designs at the same time.
The advantage of coding of variable is that coding discretizes the
search space even though the function may be continuous.
Traditional optimization methods use transition rules that are
deterministic in nature.
While genetic algorithm uses randomize operators. Randomize
operator improve the search space in an adaptive manner.

There are three important aspects of Genetic Algorithm are-

1. Definition of objective function.

2. Definition and implementation of genetic representation.

3. Definition and implementation of Genetic operators.

Advantages of Genetic Algorithm (GA) :-

1. It shows simplicity.
2. Ease of operation.
3. Minimal requirement.
4. Global perspective.
5. It does not guarantee to find global minimum solutions but acceptably good
solutions to “acceptably quickly “.

Flowchart of Genetic Algorithm (GA) :-

Genetic Algorithm Steps :-

1. BEGIN
2. Create initial population ;
3. Compute fitness of each individuals ;
4. WHILE NOT finished DO Loop
5. BEGIN
6. Select individuals from old generation for mating ;
7. Create offspring by applying crossover or mutation to the selected individuals ;
8. Compute fitness of new individuals ;
9. Kill old individuals to make a room for new chromosomes and insert offspring in
the new
generation ;
10. If population has converged
11. Then fitness=TRUE ;
12. END
13. END
 EXPERIMENT: 6

DESCRIPTION: STUDY AND ANALYSIS OF CPN.

Counter Propagation Network (CPN) :-

Counter Propagation Network (CPN) is not as general as Back

Propagation. But it provides a solution for those application that cannot tolerate long
training time.
It is a combination of two well known algorithm i.e. Self Organizing
Map of Kohonen and Grossberg outstar. The training process associates the input
vector with the corresponding output vector.
These vectors may be binary or continuous. It has the generalization
capability. The generalization capability of network allows it to produce a correct
output even when it is given an input vector i.e. partially incomplete or partially
incorrect.
CPN is used to compress the data before sending and decompress the
data before receiving. With the help of these network we can compress audio data
upto 100:1 and video data upto 10:1.

Architecture Of CPN ->

It consist of one input layer, one Kohonen layer and one Grossberg
layer.
All the units of input layer are fully interconnected by weights to the
units of Kohonen layer.
Similarly, all the units of Kohonen layer are fully interconnected by
weights to the Grossberg layer.
It works in two mode.

1. Normal Mode ->

When we apply the input vector and get the desired output.

3. Training Mode ->

Apply input vector and modify the network weights so that, we
get the desired output.
Advantages of CPN ->

4. It is simple and it forms a good statistical model of its input vector

environment.

5. CPN trains rapidly. If appropriately applied, it can safe large amount of

computer time.

6. It is also useful for rapid prototyping of system, where the greater accuracy of
Back Propagation makes it the method of choice in the final version. But a
quick approximation is important so, CPN is more useful then Back
Propagation.
 EXPERIMENT: 7

DESCRIPTION: STUDY AND ANALYSIS OF FUZZY Vs CRISP LOGIC.

FUZZY LOGIC :-

Fuzzy logic is a form of many valued Logic. It deals with reasoning

i.e. approximate or inexact.
Fuzzy logic have been extended to handle the concept of partial truth
where the truth value will be in the range between completely true and completely
false.
Fuzzy logic began with the 1965 by Lotfi Zadeh. Fuzzy logic has been
applied to many fields from control theory to artificial intelligence.
CRISP LOGIC :-

In older days, Crisp logic were used to handle the problem of binary
value ie. 0 and 1. Crisp logic is also known as traditional, conventional
or binary logic.
Crisp logic have two valued logic first is true and other is false. Crisp
logic is based on the reasoning which is exact and fixed. It is based on the logic of
completely true and completely false.
We can defined completely true as one (1) and completely false as zero
(0).
 EXPERIMENT:7

DESCRIPTION: Write a program of Perceptron Training Algorithm.

Algorithm
Start with a randomly chosen weight vector w0;
Let k=1;
While these exists input vector that are misclassified by: Wk-1 do
Let i be a misclassified input vector
Let Xk=class(ij)ij, impling that Wk-1.Xk<0
Update the weight vector to Wk= Wk-1 + nXk;
increment k;
End while;

Program
#include<iostream.h>
#include<conio.h>

Void main( )
{
clrscr( );
int in[3],d,w[3],a=0;
for(int i=0;i<3,i++)
{
cout<<”\n initialize the weight vector w”<<i;
cin>>w[i]
}
for(i=0;i<3:i++}
{
cout<<”\n enter the input vector i”<<i;
cin>>in[i];
}
cout<<”\n enter the desined output”;
cin>>d;
int ans=1;
while(ans= = 1)
{
for (a= 0, i==0;i<3;i++)
{
a = a + w[i] * in[i];
}
clrscr( );
 cout<<”\n desired output is”<<d;
cout<<”\n actual output is “<<a;
int e;
e=d-a;
cout<<”\n error is “<<e;
cout<<”\n press 1 to adjust weight else 0”;
cin>>ans;
if (e<0)
{
for(i=0;i<3;i++)
{
w[i]=w[i]-1;
}
}
else if (e>0)
{
for(i=0;i<3:i++)
{
w[i]=w[i]+1;
}
}
getch( );
}

OUTPUT:
 EXPERIMENT:8

DESCRIPTION: Write a program to implement Hebb’s rule

#include<<iostream.h>>
#include<<conio.h>>
void main()
{
float n,w,t,net,div,a,al;
cout<<”consider o single neuron percetron with a single i/p”;
cin>>w;
cout<<”enter the learning cofficient”;
cin>>d;
for (i=0;i<10;i++)
{
net = x+w;
if(wt<0)
a=0;
else
a=1;
div=at+a+w;
w=w+div;
cout<<”i+1 in fraction are i”<<a<<”change in weight”<<dw<<”adjustment at=”<<w;
}
}

OUTPUT:
 EXPERIMENT :9

DESCRIPTION: Write a program to implement of delta rule.

#include<<iostream.h>>
#include<<conio.h>>
void main()
{
clrscr( );
float input[3],d,weight[3],delta;
for(int i=0;i < 3 ; i++)
{
cout<<”\n initilize weight vector “<<i<<”\t”;
cin>>input[i];
}
cout<<””\n enter the desired output\t”;
cin>>d;
do
{
del=d-a;
if(del<0)
for(i=0 ;i<3 ;i++)
w[i]=w[i]-input[i];
else if(del>0)
for(i=0;i<3;i++)
weight[i]=weight[i]+input[i];
 for(i=0;i<3;i++)
{
val[i]=del*input[i];
weight[+1]=weight[i]+val[i];
}
cout<<”\value of delta is “<<del;
cout<<”\n weight have been adjusted”;
}while(del ≠ 0)
if(del=0)
cout<<”\n output is correct”;
}

OUTPUT
 EXPRIMENT :10

DESCRIPTION: Write a program for Back Propagation Algorithm

# include <iostream.h>
#include <conio.h>
void main ()
{
int i ;
float delta, com, coeff = 0.1;
struct input
{
float val,out,wo, wi;
int top;
} s[3] ;

cout<< “\n Enter the i/p value to target o/p” << “\t”;
for (i=0; i<3 ; i++)
cin>> s [i], val>> s[i], top);
i = 0;
do
{
if (i = = 0)
{
W0 = -1.0;
W1 = -0.3;
}
else
{
W0 = del [i - 1], W0 ;
W1 = del [i - 1] , Wi ;
}
del [i]. aop = w0 + (wi * del [i]. val);
del [i].out = del [i]. aop);
delta = (top – del [i]. out) * del [i].out * (1 – del [i].out);
corr = coeff * delta * del [i].[out];
del [i].w0 = w1 + corr;
del [i]. w1 = w1 + corr;
i++;
}While ( i ! = 3)
cout<< “VALUE”<<”Target”<<”Actual”<<”w0” <<”w1”<<’\n;
for (i=0; i=3; i++)
{
cout<< s [i].val<< s[i].top<<s[i].out << s[i]. w0<< s[i]. w1;
cout<< “\n”;
}
getch ();
}
OUTPUT
 EXPERIMENT :11

DESCRIPTION: Write a program to implement logic gates.

#include <iostream>
int main()
{
char menu; //Menu control variable
int result; //final output variable
int dataValue1;
int dataValue2;
cout << "enter your Boolean operator code: (A,O,N,X): ";
cin >> menu;
switch (menu) //Menu control variable
{
case 'A':
cout << "Enter first Boolean value:";
cin >> dataValue1;
cout << "Enter second Boolean value:";
cin >> dataValue2;
if(dataValue1 == 1 && dataValue2 == 1)
 {
result = 1;
}
else
{
result = 0;
}
cout << "show result:" << result;
break;
case 'O':
cout << "Enter first Boolean value:";
cin >> dataValue1;
cout << "Enter second Boolean value:";
cin >> dataValue2;
if(dataValue1 == 1 || dataValue2 == 1)
{
result = 1;
}else
{
result = 0;
}
cout << "show result:" << result;
break;
case 'N':
cout << "Enter first Boolean value:";
cin >> dataValue1;
result = !dataValue1;
cout << "show result:" << result;
break;
case 'X':
cout << "Enter first Boolean value:";
cin >> dataValue1;
cout << "Enter second Boolean value:";
cin >> dataValue2;
if(dataValue1 = !dataValue1)
{
result = 1;
}else
{
result = 0;
}
cout << "show result:" << result;
break;
default:
result = 0;
break;
}//end switch

cin.ignore(2);
return 0;
}//end main

OUTPUT